About
I help startups build the right product, the right way. As a Fractional CTO, I provide…
Services
Articles by Aleksandr
Activity
5K followers
-
Aleksandr Protsiuk posted thisFive years from now, someone will build a $100M company alone. Not a small team. One founder. 100 AI agents running in parallel - shipping features, handling support tickets, writing docs, running growth experiments, monitoring prod. The bottleneck won't be how fast you can execute. It'll be judgment. Mental model. Taste. Who understands the market deeply enough to ask the right questions. Who knows which 10% of features actually matter. Who says no to the other 90%. The solo founder era is coming. Not because humans got smarter - because execution got cheap. Happy Friday. The future is weird and I love it.
-
Aleksandr Protsiuk shared thisLast quarter I reviewed five startups that wanted to add AI to their product. Four of them had the same problem. Not with the AI part - with everything underneath it. Latency that was fine at 100 users breaks visibly at 1,000 when you add a model call in the critical path. Data pipelines that worked for analytics don't work for inference. Auth layers that were never designed for async callbacks start failing in ways that are hard to reproduce. The founders all said some version of the same thing: "We thought we were adding a feature. We didn't realize we were changing the architecture." Here's the pattern I see: - AI works great in the demo because demos are synchronous, clean, and forgiving - AI breaks in production because production is none of those things - The bottleneck is never the model - it's the ten-year-old database query running behind it Before you integrate AI into your product, you need answers to three questions: 1. What's your p99 latency on the endpoint you're adding this to? 2. Do you have observability on that path - not just logs, but traces? 3. If the model call fails, does your product degrade gracefully or go down? If you don't know the answers, you're not ready to ship AI. You're ready to ship a demo. The gap between those two things is where I spend most of my time with founders. If you're planning an AI integration and want a second opinion before you build: https://lnkd.in/gV9V5ukm #fractionalcto #aiproducts #startuptechnology
-
Aleksandr Protsiuk posted thisInvestors asked a founder I work with to share their architecture diagram. He sent a Notion page from 2018 with a box labeled "main service" pointing to something that no longer existed. They passed. Not because the tech was bad. Because nobody looked like they were steering. What VCs actually care about in tech due diligence - from watching enough of these go sideways: Who made the architectural decisions. Not who wrote the code. Who decided. "Our agency handled that" is a red flag - not because agencies are bad, but because you don't have anyone to walk investors through the reasoning. Scalability is a trick question. They don't expect you to handle Netflix traffic at $2M ARR. They want to know if you understand your own bottlenecks. "We'd need to think about sharding around 5M events/day" beats "our CTO says we're fine." Known debt vs. hidden debt. Every codebase has shortcuts. But I once found a hardcoded API key in a client's repo the week before their Series A. They didn't know it was there. Round got delayed 6 weeks. If I can find it in an hour, the investor's technical advisor can too. Can your current team build what's on the roadmap. "We need 2 senior engineers to do this right" is a fundable answer. "We're good" when you're clearly not - that kills credibility. The founders who do best in these reviews aren't the ones with the cleanest code. They're the ones who've sat with the hard questions long enough to know what they don't know. Dry run before your raise - link in comments.
-
Aleksandr Protsiuk shared thisJust got my pass for Stripe Sessions 2026—looking forward to meeting up in San Francisco in April! stripesessions.comStripe Sessions 2026 | The global internet economy conferenceStripe Sessions 2026 | The global internet economy conference
-
Aleksandr Protsiuk posted thisI've been deep in the architecture of how AI agents interact with each other. And it's changing how I think about teams. A single agent is impressive. Give it a task - get a result. But that's like having 100 people working in isolation. Productive, sure. But not even close to what small teams pull off when they share context and build on each other's work. Same thing with agents. Give them a group, a reporting structure, a way to communicate - something clicks. They start assigning tasks to each other, reviewing each other's code, moving tickets on a board. Staying in their lane. At some point you stop managing agents and start managing an organization. You're designing the structure, deciding where to draw lines, translating business problems into something this machine can execute. Which is basically the CTO job. Doesn't matter if your org chart has people or agents in it. Weird realization after 15 years of writing code. But this might be the most interesting engineering problem I've worked on.
-
Aleksandr Protsiuk posted thisThe decisions you make in month 2 of your startup will haunt you in month 12. I've watched it happen more times than I can count. Someone picks the wrong database because a blog post said so. Hires a full-stack dev who can't actually ship. Tells themselves "we'll refactor later" knowing damn well they won't. And lately there's a new one - founders vibe-coding their MVP with AI, moving fast, and having zero idea what's actually in their codebase. It works until it doesn't. I spent 15 years building products for startups. Shipped over 200. Made plenty of my own mistakes along the way. Now I work as a Fractional CTO - basically the technical co-founder you bring in without the equity conversation. Most of what I do comes down to four things: reviewing architecture before you're locked in, helping you hire engineers who actually deliver, prepping you for investor meetings so technical questions don't throw you off, and figuring out build vs buy before you waste six months on the wrong path. If you vibe-coded your way to an MVP and need someone to tell you what's solid and what's a ticking time bomb - that's literally my job now. I do one-time audits (a week, honest breakdown of your stack and team), ongoing retainers (code reviews, sprint planning, hiring calls), and investor-ready tech reviews for founders raising their next round. Or just a single 60-minute call if you have one specific problem. 200+ products shipped. Stanford '23. API World 2024 Hackathon winner. Based in San Francisco. No $300k salary. No 4-year commitment. Just a guy in San Francisco who's seen every way this can go wrong. Link to book a call in the comments.
-
Aleksandr Protsiuk posted thisLast month I scraped 870 one-to-three-star reviews of Evernote, Notion, Bear, and Obsidian. I expected people to complain about missing AI features. They did not. The most common complaint across all four apps? Price hikes. Evernote users especially. One person called it "ransomware" after the price jumped from $50 to $250 a year and they could not access their own notes without paying. But the one that stuck with me was a Notion user who wrote: "they ruined it by becoming a slop-first app, note-taker second." That line keeps rattling around in my head because it captures what is happening to the whole category. Every note-taking app is in an AI arms race. Nobody stopped to ask if users wanted it. 65 of those 870 reviews specifically said "stop pushing AI on me." This shaped how we built ContextorAI. The AI is there but you do not see it until you search for something you wrote months ago. No AI toolbar. No generate button in your face. Just retrieval that actually works. I keep wondering if the industry is about to learn an expensive lesson about the difference between AI that helps and AI that annoys. What app did you leave recently, and what finally broke you? #AI #NoteTaking
-
Aleksandr Protsiuk posted thisMy teenage son keeps all his notes in Telegram Saved Messages. No folders. No tags. Just one long chat with himself. I have 2,649 notes in Apple Notes. I built a whole system -- folders by project, tags by topic, smart folders. The works. And last Tuesday I spent 8 minutes looking for a client's phone number I knew I saved somewhere. My son would have found it in 3 seconds. He just scrolls up. That bugged me for days. I kept thinking -- we are building ContextorAI with knowledge graphs, encryption layers, rich text editors. Real features solving real problems. But somewhere along the way I forgot the actual job. The job is not "organize your knowledge." The job is "find the thing when you need it." I almost cut three features from our roadmap that week. Ended up keeping them, but reframing everything around retrieval instead of storage. The AI answers your question. You do not need to remember where you put something. Funny thing is, my son still will not switch. "Dad, Telegram works fine." He is right. The best system is the one you do not think about. What tool do you use that you would never recommend to anyone but secretly works perfectly for you? #NoteTaking #AI
-
Aleksandr Protsiuk posted thisAnthropic interviewed 81,000 Claude users across 159 countries. The largest qualitative AI study ever done. What people want from AI: - 32% want to speed up work - 17% want a "cognitive partner" for brainstorming - 13.5% want life management, less cognitive load - 11% want their time back from routine tasks What people fear: - 27% worry about hallucinations and inaccuracy - 22% fear job loss - 16% fear losing their own skills The part that stuck with me: hope and fear live in the same people. Not "optimists vs pessimists." The same person wants AI to reduce cognitive load - and is afraid of forgetting how to think. We're building ContextorAI for exactly this gap. AI that only works with your data. It doesn't generate - it finds. It doesn't think for you - it helps you remember what you already know.
Aleksandr Protsiuk commented on a post
1d
$1M to $100M in 18 months in legal AI also means compliance and data architecture scaled 100x in parallel - legal software has a higher bar for audit trails and data residency than most B2B SaaS. curious whether the infra decisions made at $1M were still holding at $100M or there was a major re-architecture along the way.
Aleksandr Protsiuk commented on a post
1d
the CTO investing in proper DB/infra separation early is exactly the right call - most teams do it at the API layer and call it done, then wonder why their internal dogfooding gives different results than customer production. that investment being made "very early on" is what makes the dogfooding real.
Experience & Education
-
Self-employed
View Aleksandr’s full experience
See their title, tenure and more.
or
Projects
Honors & Awards
-
API World 2024 Hackathon Winner
API World 2024
Zavod-IT Founder Triumphs at API World 2024 Hackathon in Silicon Valley – Leading the Way in API Innovation
Languages
-
English
Full professional proficiency
-
Russian
Native or bilingual proficiency
Other similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content